This document summarizes a study on prosody analysis for speech signals across different emotions. It discusses how prosody relates to features like pitch, duration, jitter and shimmer. The study analyzed speech recordings in Odia language uttered with five emotions - anger, love, neutral, sadness and calmness. Acoustic measurements were made on extracted vowels to analyze impact of emotions on parameters like duration, fundamental frequency, jitter and shimmer. The results provide insights into how prosody conveys emotional information in speech.
IJCER (www.ijceronline.com) International Journal of computational Engineering research
1. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
A Study on Prosody Analysis
Padmalaya Pattnaik [1], Shreela Dash [2]
(Asst.Prof, C.V. Raman College of Engineering, Bhubaneswar, Odisha, India)
Abstract:
Speech can be described as an act of producing voice through the use of the vocal folds and vocal apparatus to create a
linguistic act designed to convey information. Linguists classify the speech sounds used in a language into a number of
abstract categories called phonemes. Phonemes are abstract categories, which allow us to group together subsets of speech
sounds. Speech signals carry different features, which need detailed study across gender for making a standard
database of different linguistic, & paralinguistic factors. Prosodic phenomena are specific to spoken language. They
concern the way in which speech sounds are acoustically realized: how long they are, how high and how loud. Such
acoustic modulations are used by human speakers to express a variety of linguistic or paralinguistic features, from stress
and syntactic boundaries, to focus and emphasis or pragmatical and emotional attitudes. Linguistics and speech technology
have approached prosody from a variety of points of view, so that a precise definition of the scope of prosodic research is
not easy. A main distinction can be drawn between acoustic-phonetic analyses of prosody and more abstract, linguistic,
phonological approaches When people interact with others they convey emotions. Emotions play a vital role in any kind
of decision in affective, social or business area. The emotions are manifested in verbal, facial expressions but also in
written texts. The objective of this study is to verify the impact of various emotional states on speech prosody analysis.
Keywords: Duration, Emotion, Jitter, Prosody, Shimmer
1. Introduction
No language is produced in a smooth, unvarying stream. Rather, the speech has perceptible breaks and clumps. For
example, we can perceive an utterance as composed of words, and these words can be perceived as composed of syllables,
which are composed of individual sounds. At a higher level, some words seem to be more closely grouped with adjacent
words: we call these groups phrases. These phrases can be grouped together to form larger phrases, which may be grouped
to form sentences, paragraphs, and complete discourses. These observations raise the questions of how many such
constituents there are and how they are best defined. A fundamental characteristic of spoken language is the relation
between the continuous flow of sounds on the one hand, and the existence of structured patterns within this continuum on
the other hand. In this respect, spoken language is related to many other natural and man-made phenomena, which are
characterized not only by their typically flowing nature but also by the fact that they are structured into distinct units such
as waves and measures. Prosodic phonology is a theory of the way in which the flow of speech is organized into a finite set
of phonological units. It is also, however, a theory of interactions between phonology and the components of the grammar.
Although many speech interfaces are already available, the need is for speech interfaces in local Indian languages.
Application specific Indian language speech recognition systems are required to make computer aided teaching, a reality in
rural schools. This paper presents the preliminary work done to demonstrate the relevance of an Oriya Continuous Speech
Recognition System in primary education. Automatic speech recognition has progressed tremendously in the last two
decades. There are several commercial Automatic Speech Recognition (ASR) systems developed, the most popular among
them are Dragon Naturally Speaking, IBM Via voice and Microsoft SAPI. Speech is a complex waveform containing
verbal (e.g. phoneme, syllable, and word) and nonverbal (e.g. speaker identity, emotional state, and tone) information. Both
the verbal and nonverbal aspects of speech are extremely important in interpersonal communication and human-machine
interaction. Each spoken word is created using the phonetic combination of a set of vowel semivowel and consonant
speech sound units. Different stress is applied by vocal cord of a person for particular emotion. The increased muscle
tension of the vocal cords and vocal tract can directly or indirectly and adversely affect the quality of speech. We use
emotions to express and communicate our feelings in everyday life. Our experience as speakers as well as listeners
tells us that the interpretation of meaning orintention of a spoken utterance can be affected by the emotions that area
expressed and felt.
2. Literature
According to the classic definition, prosody has to do with speech features whose domain is not a single phonetic segment,
but larger units of more than one segment, possibly whole sentences or even longer utterances. Consequently, prosodic
phenomena are often called supra-segmentals. They appear to be used to structure the speech flow and are perceived as
stress or accentuation, or as other modifications of intonation, rhythm and loudness.
Issn 2250-3005(online) September| 2012 Page 1594
2. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
Fig: 1 Stages of oral communication
An emotion is a mental and physiological state associated with a wide variety of feelings, thoughts, and
internal (physical) or external (social) behaviors. Love, hate, courage, fear, joy, sadness, pleasure and disgust can all
be described in both psychological and physiological terms. An emotion is a psychological arousal with cognitive aspects
that depends on the specific context. According to some researcher, the emotions are cognitive processes. Emotion is a
process in which the perception of a certain set of stimuli, follows cognitive assessment which enables people to label
and identify a particular emotional state. At this point there will be an emotional physiological, behavioral and
expressive response. For example, the primordial fear, that alerts us as soon when we hear a sudden noise, allows to react
to, dangerous situations and provides instantly resources to face them as escape or close the door. The emotional stimuli
may be an event, a scene, a face, a poster, an advertising campaign. These events, as a first reaction, put on alert the
organism with somatic changes as heart rate, increase of sweat, acceleration of respiratory rhythm, rise of muscle tensions.
Emotions give an immediate response that often don't use cognitive processes and conscious elaboration and
sometimes they have an effect on cognitive aspects as concentration ability, confusion, loss, alert and so on. This is
what is asserted in evaluation theory, in which cognitive appraisal is the true cause of emotions [2]. Two factors that
emerge permanently are those related to signals of pleasure and pain and characterizing respectively the positive
and negative emotions. It’s clear that these two parameters alone are not sufficient to characterize the different emotions.
Many authors debate on primary and secondary emotions other on pure and mixed emotions, leaving the implication that
emotions can somehow be composed or added.
The systems based on the analysis of physiological response as blood pressure, heart rate, respiration
change present an initial phase where the signals are collected in configurations to be correlated with different emotional
states and a subsequently recognition basing on the measure of indicators. One of the interesting early works on the
emotions was that one of Ortony [3]. From this work, through componential analysis, other authors constructed an
exhaustive taxonomy on affective lexicon. According to Ortony, stimuli that cause emotional processes are of three basic
types: events, agents and objects corresponding to three classes of emotions: satisfied/ unsatisfied (reactions to events),
approve/disapprove(reaction to agents), appreciate/unappreciate (reaction to objects). According to Osgood [4] an
emotion consists of a set of stages: stimulus (neural and chemical changes), appraisal and action readiness. Continuing the
studies of Charles Darwin, the Canadian psychologist Paul Ekman [5] has confirmed that an important feature of
basic emotions is that they are universally expressed, by everybody in any place, time and culture, through similar
methods. Some facial expressions and the corresponding emotions are not culturally specific but universal and they
have a biological origin. Ekman, analyzed how facial expressions respond to each emotion involving the same type of
facial muscles and regardless of latitude, culture and ethnicity. This study was supported by experiments conducted with
individuals of Papua New Guinea that still live in a primitive way.
3. Emotions
Human emotions are deeply joined with the cognition. Emotions are important in social behavior and to
stimulate cognitive processes for strategies making. Emotions represent another form of language universally spoken
and understood. Identification and classification of emotions has been a research area since Charles Darwin's age. In this
section we consider facial, vocal and textual emotional expressions.
Emotion classifications of the researchers differ according to the goal of the research and the field .Also the
scientist’s opinion about the relevance of dividing different emotions is important. There is no standard list of basic
emotions. However, it is possible to define list of emotions which have usually been chosen as basic, such as:
erotic(love)(shringar), pathetic (sad) (karuNa), wrath (anger)(roudra), quietus (shAnta), normal(neutral).
The objective of this study is to analyze impact for different emotions on vowels in terms of certain
parameters for stage prosody analysis.
4. Prosody
In linguistics, prosody is the rhythm, stress, and intonation of speech. Prosody may reflect various features of the
speaker or the utterance: the emotional state of the speaker; the form of the utterance (statement, question, or command);
the presence of irony or sarcasm; emphasis, contrast, and focus; or other elements of language that may not be encoded by
grammar or choice of vocabulary. Prosody has long been studied as an important knowledge source for speech
understanding and also considered as the most significant factor of emotional expressions in speech[16]. Prosody gives
Issn 2250-3005(online) September| 2012 Page 1595
3. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
naturalness and message intelligibility to speech Emotional prosody is the expression of feelings using prosodic elements
of speech. Linguistically relevant prosodic events concur to express sentence structure: they highlight linguistic units by
marking their boundaries and suggesting their function. Several types of prosodic units (differing mainly in their scope)
have been proposed: paragraphs, sentences, intonation groups, intermediate groups, stress groups, syllables etc. Although
prosody is by definition suprasegmental, prosodic analyses take often the phoneme as their minimal unit, where to measure
rhythmical variations and locate intonation events. The family of prosodic phenomena includes the suprasegmental features
of intonation, stress, rhythm and speech rate, whose variations are relevant to express the function of the different prosodic
units: the prominent syllable in the word will be marked by stress, a falling intonation contour will mark the conclusion of
a sentence, a faster speech rate and lower intonation characterize a parenthetical phrase.Such prosodic features are
physically realized in the speech chain in terms of variations of a set of acoustic parameters. Acoustic-phonetic analyses
identify the following ‘phonetic correlates of prosody’: fundamental frequency/pitch (f0), length changes in segmental
duration, pauses, loudness, voice quality.
Prosody is the combination of voice's pitch, duration and energy variation during speech. It provides an additional
sense to the words, which is extraordinarily important in natural speech. For example, interrogative and declarative
sentences have very different prosody (especially intonation). Besides, the prosody of a sentence is one of the factors that
make a speaker seem happy, sad, angry or frightened. We can even decide from prosody if the speaker is an energetic
person or, on the contrary, a lazy one. When singing, intonation and timing evolution characterize melody. But prosody is
not only important in natural speech but also in synthetic speech. Prosody is crucial in order to achieve an acceptable
naturalness. If a TTS system does not have a good prosodic treatment, its output speech sounds completely monotonous
and, moreover, it won't be able to distinguish between sentences of different kinds. The two main parameters of prosody
are-
4.1 Intonation
In a first approximation, sounds can be classified invoiced and unvoiced sounds. Voiced sounds, unlike unvoiced ones, are
produced making the vocal chordsvibrate. These vibrations provoke some periodicitiesin the speech signal and therefore a
fundamental frequency (F0). This value is inversely proportional tothe distance between periodicities and it makesspeech
sound with higher or lowers frequency. It iscommonly called pitch. On the contrary, as unvoiced sounds do not have any
periodicities (vocal chords donot vibrate) and can be modeled as a filtered noise signal. So if we detect the pitch curve of a
speech signal it will only exist in the voiced segments. Pitch is not constant but its value changes during asentence. That is
called intonation. Thanks tointonation we can distinguish, for example, between adeclarative and an interrogative sentence
or identifyfocused words inside a sentence.
4.2 Duration
The duration of speech segments is the other main parameter of prosody. The timing structure of a sentence is extremely
important to give naturalness to speech. Phone duration depends on a great number of parameters, such as its phonetic
identity, surroundingphones, level of stress or position in the sentence or in the word. What’s more, duration of a word also
depends on its importance in the sentence. For example, a focused word will generally have longer duration.
5. Prosody Analysis
TTS systems generate speech from a text. There is a need of prosodic assignment to phones to produce high quality speech.
Once the phones to synthesize are determined, it is necessary to know the pitch and duration yet achieved the required
quality needed for most applications. This is the main reason why the tool was created: the prosodic module requires
models of prosodic patterns and these patterns have to be studied and tested before the application to TTS.The goal of the
prosodic analysis is the generation of the pitch contour[17].
Fig: 2 Speech Segment & it’s pitch contour
Issn 2250-3005(online) September| 2012 Page 1596
4. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
6. Parametric Measurements of Acoustics Signals for Prosody Analysis
Four acoustic signal features such as Duration of speech segment, Pitch, Jitter and Simmer were used to
parameterize the speech.
6.1 Duration
Utterance durations, vowel durations were measured from the corresponding label files produced by a
manual segmentation procedure. On an average, utterance durations become longer when speech is emotionally
elaborated.
6.2 Fundamental Frequency (pitch)
We calculated the pitch contours of each utterance using speech processing software. Global level statistics related
to F0 such as minimum, maximum, mean were calculated from smoothed F0 contours.
6.3 Jitter & Shimmer
Jitter and Shimmer are related to the micro-variations of the pitch and power curves. In other words,
Shimmer and Jitter are the cycle-to-cycle variations of waveform amplitudes and fundamental periods respectively. The
Jitter & Shimmer occur due to some undesirable effect in audio signal. Jitter is the period frequency displacement of
the signal from the ideal location. Shimmer is the deviation of amplitude of the signal from the ideal location.
7. Methodology and Experimentation
There are many features available that may be useful for classifying speaker affect: pitch statistics, short-time
energy, long-term power spectrum of an utterance, speaking rate, phoneme and silence durations, formant ratios, and even
the shape of the glottal waveform [8, 11, 12, 9, 10]. Studies show, that prosody is the primary indicator of a speaker's
emotional state [1, 13, 12]. We have chosen to analyze prosody as an indicator of affect since it has a well-defined and
easily measureable acoustical correlate -- the pitch contour. In order to validate the use prosody as an indicator for affect
and to experiment with real speech, we need to address two problems: First, and perhaps most difficult, is the task of
obtaining a speech corpus containing utterances that are truly representative of an affect. Second, what exactly are the
useful features of the pitch contour in classifying affect? Especially as many factors influence the prosodic structure of an
utterance and only one of these is speaker’s emotional state [6, 7, 9].
The data analyzed in this study were collected from semi-professional actors and actress and consists of 30
unique Odia language sentences that are suitable to be uttered with any of the five emotions i.e., roudra, shringar,
shanta, karuna, and neutral. Some example sentences are “Jayanta jagi rakhhi kaatha barta kaara”, “Babu chuata mora
tinni dina hela khaaini”. The recordings were made in a noise free room using microphone. For the study, samples have
been taken from three male & three female speakers. The utterances are recorded at the bit rate of 22,050Hz. The Vowels
are extracted from the words consisting of 3 parts i.e.CV, V, VC. CV stands for Consonant to Vowel transition, V for
steady state vowel, VC for Vowel to Consonant transition.
8. Experimental Results
For experimental analysis data samples were created from recordings of male and female speakers in
various emotions (mood). Vowels are then extracted and stored in a database for analysis. From this database after
analysis the result of the utterance. Duration, fundamental frequency & variations of pitch of every vowel are measured
and compared to give following results.
8.1 Duration
It is observed from the duration Table 1 that speech associated with the emotion “love( Shringar)”has higher
duration with the vowel /i/ gets elongated both for male and female speakers where as the emotion “sad (karuna)” for
male speakers vowel(/a/,/i/) gets elongated whereas for female (/i/,/u/) gets elongated.
Issn 2250-3005(online) September| 2012 Page 1597
5. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
8.2 Fundamental Frequency
Figure3 shows the analysis result that the mean pitch for male speaker associated with emotion karuna for vowel
/i/ has dominance where as in female the speech associated with emotion shringar for vowel /a/ plays dominant role.
Fig: 3 mean pitch of vowels of speakers in different emotion
8.3 Jitter
Figure 4 show that the emotion “anger” of vowel /i/ is having a dominant role in jitter for male speaker. The jitter
value for female speaker has dominant vowel /u/ for the emotion “love”.
Fig:4 jitter of vowels of speakers in different emotion
8.4 Shimmer
It is observed from figure 5 that the shimmer in the emotion “anger” of vowel /o/ has dominant role for males &
for female in emotion “anger” of vowel /i/ has dominant role.
Fig:5 shimmer of vowels of speakers in different emotion
Issn 2250-3005(online) September| 2012 Page 1598
6. International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 5
9. Conclusion
The great importance and complexity of prosody in speech makes this subject an important area of research in
speech synthesis applications In this study, we investigate acoustic properties of speech prosody associated with five
different emotions (love (shringar)), pathetic (sad) (karuNa), wrath (anger) (roudra), quietus (shAnta),normal
(neutral) intentionally expressed in speech by male and female speakers. Results show speech associated with love
(shringar) and sad (karuna) emotions are characterized by longer utterance duration, and higher pitch. However we
observed that for jitter anger or love has dominance over others, whereas for shimmer the emotion anger plays a vital
role. Future works of my research are the following. We have to collect synthetic speech and put emotion labels on them.
We have to reconsider how to estimate emotion in speech using parallel programming.
References
[1] P. Olivier and J. Wallace, Digital technologies and the emotional family, International Journal of Human
Computer Studies, 67 (2), 2009, 204-214.
[2] W. L. Jarrold, Towards a theory of affective mind: computationally modeling the generativity of goal
appraisal, Ph.D. diss., University of Texas, Austin, 2004.
[3] C. G. Ortony and A. Collins, The cognitive structure of emotions, (Cambridge University Press: New York,
1990).
[4] M. M. C.E. Osgood and W.H. May, Cross-cultural Universals of Affective Meaning, (Urbana Chaimpaigne:
University of Illinois Press, 1975).
[5] E. Paul, Emotions Revealed: Recognizing Faces and Feelings to Improve Communication and Emotional Life,
(NY: OWL Books, 2007).
[6] A. Ichikawa and S. Sato, Some prosodical characteristics in spontaneous spoken dialogue, International
Conference on Spoken Language Processing, v. 1, 1994, 147-150.
[7] R. Collier, A comment of the prediction of prosody, in G. Bailly, C. Benoit, and T.R. Sawallis (Ed.), Talking
Machines: Theories, Models, and Designs, (Amsterdam: Elsevier Science Publishers, 1992).
[8] H. Kuwabara and Y. Sagisaka, Acoustic characteristics of speaker individuality: Control and conversion, Speech
Communication, 16(2), 1995, 165-173.
[9] K. Cummings and M. Clements, Analysis of the glottal excitation of emotionally styled and stressed speech,
Journal of the Acoustical Society of America, 98(1), 1995, 88-98.
[10] D. Roy and A. Pentland, Automatic spoken affect classification and analysis, Proceedings of the 2nd International
Conference on Automatic Face and Gesture Recognition, 1996, 363-367.
[11] A. Protopapas and P. Lieberman, Fundamental frequency of phonation and perceived emotional stress, Journal of
Acoustical Society of America, 101(4), 1997, 2267-2277.
[12] X.Arputha Rathina and K.M.Mehata, Basic analysis on prosodic features in emotional speech, nternational
Journal of Computer Science, Engineering and Applications (IJCSEA) , Vol.2, No.4, August 2012,99-107
[13] D. Hirst, Prediction of prosody: An overview, in G. Bailly, C. Benoit, and T.R. Sawallis (Ed.), Talking Machines:
Theories, Models, and Designs, (Amsterdam: Elsevier Science Publishers, 1992).
[14] L. Rabiner and R. Shafer, Digital Processing of Speech Signals, (New York: Wiley and Sons, 1978)
[15] Wavesurfer, http://www.speech.kth.se/wavesurfer/
[16] Masaki Kuremastsu et eal, An extraction of emotion in human speech using speech synthesize and
classifiers foe each emotion , International journal of circuits systems and signal processing, 2008.
[17] J .L. Navarro, I. Esquerra, A Time-FrequencyApproach to Epoch Detection, Proceedings ofEurospeech’95, 405-
408, Madrid 1995.
Issn 2250-3005(online) September| 2012 Page 1599